Texas Memory Systems goes Texas, hobbles together RamSan-620 5TB SSD "drive"
Way to play to your stereotypes, Texas. Texas Memory Systems just announced what it claims is the largest SLC SSD on the market, with 5TB of single level cell (the good kind) flash memory spread across a 2U rack shelf. The drive can handle 250,000 sustained I/Os per second, churn through 3GB of data a second, and has 80 microsecond write latency. Texas Memory Systems claims that for similar performance from a HDD setup you'd need half a million dollars and consume 20x the power. TMS calls the system "affordable for mainstream IT shops" but hasn't mentioned a price or release date just yet. We'll take two.
[Via Channel Register]
[Via Channel Register]



















i see massive porn archival potential here
Obviously you're not a golfer
One, please.
yeah, straight into the desktop ;)
ELEVENTY MILLION DOLLARS!
Yeee Hah! All you need now are cattle horns mounted on the front...
I'll take the gnarly looking case and trippy looking LEDs and fill it up with HDDs - this thing looks sweet.
There is no theoretical way you'd ever need this kind of throughput unless your DBA is a total fuckstick. You're _going to_ run in to a bottleneck elsewhere before you get to client presentation.
EvE online has 1 and i think they bought another (another model).
Not necessarily. Depends how heavy the concurrent usage is. I remember reading that EVE online uses a bunch of these and it has improved their performance significantly.
EVE actually has multiple RAMSAN 400s and a new RAMSAN 500 (if i recall correctly). The addition of the newest RAMSAN and the pending Infiniband project should eliminate most of the lag that EVE has seen since it now regularly hits 50,000 concurrent users. That's a CRAP TON of data that has to be moved around at once.
Will you ever need one at home? Never. Is it more than most businesses will ever need? Probably. Is there an existing customer base out there for the product? Definately
If your database doesn't fit in RAM, surely lots of access patterns might benefit from this kind of throughput?
The system I work on has a database that largely (not actually, but the important bits) fits in 64GB RAM, so we probably wouldn't benefit significantly from faster disk. But if the database weren't in RAM... I know caching is great, but it only takes you so far.
(moving pinky finger to the side of the mouth) One... Hundred... BILLION DOLLARS!
This needs frickin' lasers.
It already has the lasers, it just needs sharks.
So when is Engadget going to be giving one of these away as a recession antidote?
Hard drives are nice because in most raid configs they are hot swappable and easily replaced. How do you replace bad memory banks with this? I am assuming it is just going to be a bunch of ECC FBDDR2 Dimms running at PC 5300. Where is the memory mirroring/parity?
What happens to my data when a tornado knocks out power for 3 days? Can I recover critical data in the case of flooding?
I don't understand why we aren't just using hundreds of 32GB CF (the fast kind you use in a cannon DLSR) in RAID 5 with a couple GB of RAM like this set up as a battery operated write back cache in case of power outage.
TMS makes their own memory - but yea, it's basically standard RAM. It is mirrored / ECC'd to minimize loss due to memory errors. They do have hard drives in the system for redundancy in case of a major failure of the RAM. For power loss, all of the memory/HD's are battery backed so that there is time to write to disk and shut down nicely. If there is a tornado or flood, it wont matter whether it's HD's or RAM holding the data - they'll short circuit and die - that's why you always need offsite copies.
Am I right in thinking if a hard drive has been drowned / badly dinged etc you can still get the data off if you *really* want to? (With v. specialized tools of course)
As the article states, it uses SLC flash memory, not RAM.
At the moment, SSD has been all but unfulfilled claims.
I concur.
220k for the 5tb version. 88k for the 2tb.
Well... if you cannot do your business without this kind of performance, then it's definitely affordable.
But we run full 2K uncompressed files on our Stone system on much cheaper HDD raids without problems, so, to us, this doesn't make much sense.
Since it goes "Texas" is it going to threaten to secede from server when the CPU makes too many read requests?
hob·ble: To walk or move along haltingly or with difficulty; limp.
cob·ble: To put together clumsily (ex: cobbled a plan together at the last minute)
Which one were you going for there?
LOL, i totally read that as cobbles!
@Brian - Thank you, that title had me perplexed.
what bus has 3GB throughput? even optical is only like 10Gb/s
Also someone said that on a Raid0 SATA rig your throughput is higher then the 3Gb/s cap by the bus; doesn't make sense to me, but is that true?
The PCI-e 2.0 bus is 5Gb/s per lane (e.g. HBA typically uses 4-16 lanes).
The 3Gb/s limit on SATAII is per lane (e.g. mini-SAS cable has 4 lanes).
* 4-Gigabit Fibre Channel
* 2 ports standard; up to 8 ports available
* Supports point-to-point and switched fabric topologies
* Interoperable with Fibre Channel Host Bus Adaptors, switches, and operating systems
"The PCI-e 2.0 bus is 5Gb/s per lane (e.g. HBA typically uses 4-16 lanes)."
PCI-e 2.0 is 500MB/s per lane not 5Gb/s (read: 625MB/s)
http://en.wikipedia.org/wiki/PCI-e
Well, large external drive arrays (SAS /RAIDs).often connect using Infiniband cables.
Infiniband itself can scale all the way to 96Gbps using multiple channels..
This thing is nothing! Check out the RamSan-5000
Freakin 20TB of SLC!
http://www.ramsan.com/products/ramsan-5000.htm
* 1,000,000 random I/Os per second sustained (reads from Flash)
* up to 20TB Flash RAID
* up to 640GB DDR Cache
* 20 GB/s sustained bandwidth (to Flash)
* Full array of hardware redundancy to ensure availability.